Rival: held-up PU lift is a small-field summer artefact

Reasoning

Why this hypothesis exists: what the lab saw, the question it asked, the mechanism it proposed, what it predicted, and the result that would prove it wrong. Written before the test was run.

1. Observed
   • H-6c3baaba (Rival: the held-up PU lift sits in lone pull-ups, not attritional race) pointed the predicted way but failed the bar: A/E 1.28 on 530 bets (p 0.106), control 0.98. O-435140b6
   • H-6c3baaba (Rival: the held-up PU lift sits in lone pull-ups, not attritional race): the period 'after 2026-04-27' carries the effect (A/E 1.69) while the rest score 2025-10-01 to 2026-04-27 0.94. O-abea699b
2. AskedIs the A/E lift a spring/summer artefact, given the periods went 1.07 then 1.41? answered
3. MechanismSummer jumps cards have small fields, where a held-up horse that pulled up faces few rivals and is easier to place. The lift may be field size, not anything the horse kept back.
4. Why the market misses itIn small fields the bookmakers' overround is spread across fewer runners, and the P makes the horse an easy one to lay. The price may be too long simply because there are few genuine alternatives.
5. Predictedpositive, A/E 1.30+ on 350+ bets if the field-size rival holds
6. Falsified ifA/E at or below the parent's 1.198 in small fields, which would rule out field size as the carrier of the lift
7. Rule sketchlast_pulled_up = 1 AND last_run_style IN ('held_up', 'midfield') AND field_size <= 9
8. question id

The question's answer so far: Partly. The late skew is not field size (the small-field cut failed), and the changed-going cut beat par before late April too (1.17 on 635). The lift is not purely a spring/summer artefact, though the later period is still stronger (1.63).

Mechanism and rationale

Summer jumps cards have small fields, where a held-up horse that pulled up faces few rivals and is easier to place. The lift may be field size, not anything the horse kept back.

The rule

last_pulled_up = 1 AND last_run_style IN ('held_up', 'midfield') AND field_size <= 9

Reads: last_pulled_up, last_run_style, field_size

Pre-registration

Spec hash (sha256 of the canonical pre-spec)

4eda4df4d6977766cabca4f16b8b0bd5251f128bd3b092c5294eee8794e1d4d9

Frozen

not frozen

The definition can still change.

Confounders

• season: small fields cluster in the summer jumps period (after late April), the same window where the lone-PU lift sat, so field size and season are tangled and need a check by period
• going_group: summer ground is faster, which may suit horses that pulled up on soft winter ground
• sp_decimal: small fields shorten every price; the lift may sit in a band of prices, and 16/1+ has failed in every split
• race_class: summer cards are lower class, so a class drop may stand in for field size
• race-level pricing: overround per runner is different in small fields, which the same-race control picks up
• overlap with the long-gap split: horses coming back after a break in summer may be the same runners

Control

{
  "description": "The same held-up/midfield pull-up pool in fields of 10 or more runners. If field size carries the lift, the small-field rule should clearly beat this group (expected near par, roughly 1.0 to 1.1). If the large-field group shows a similar or bigger lift, the field-size rival is ruled out. The other runners in the same small-field races should show about par (roughly 0.95 to 1.0). If they also beat their prices, the lift is a property of small-field pricing (overround spread) and not of the horse.",
  "rule": "last_pulled_up = 1 AND last_run_style IN ('held_up', 'midfield') AND field_size >= 10"
}

Discovery result

fail broad rule · 17 Oct, 01:46

Bets

595

54 winners

A/E

1.09

95%: 0.82 to 1.42

SP return

-42.7%

level stakes

Betfair SP return

-27.5%

592 bets with a BSP

Strike rate 9.1%. p = 0.28. Same-race control: the other runners in these races had A/E 0.99, so the selection beat its own races. Zone: discovery. Evaluator eval-1.0, data 6a4e98be18e83933

Falsification

No attacks yet. Only promising ideas are attacked.

Sealed holdout (vault)

Not queried. Only a frozen idea can spend its one query against the 2020-25 holdout.

Forward record

No forward record. Shadow tracking starts after the vault and counts only races run after the freeze.

Lineage

Agent runs

• designer v1ok · 47s · 17 Oct, 01:44

  Output (schema-checked)

  {
    "rule": "last_pulled_up = 1 AND last_run_style IN ('held_up', 'midfield') AND field_size <= 9",
    "refinements": [
      {
        "label": "16/1 or shorter (no held-up PU split has beaten par at 16/1+)",
        "rule": "last_pulled_up = 1 AND last_run_style IN ('held_up', 'midfield') AND field_size <= 9 AND sp_decimal <= 17"
      },
      {
        "label": "jumps races only (the mechanism is about summer jumps cards)",
        "rule": "last_pulled_up = 1 AND last_run_style IN ('held_up', 'midfield') AND field_size <= 9 AND race_type IN ('Hurdle', 'Chase')"
      }
    ],
    "control": {
      "description": "The same held-up/midfield pull-up pool in fields of 10 or more runners. If field size carries the lift, the small-field rule should clearly beat this group (expected near par, roughly 1.0 to 1.1). If the large-field group shows a similar or bigger lift, the field-size rival is ruled out. The other runners in the same small-field races should show about par (roughly 0.95 to 1.0). If they also beat their prices, the lift is a property of small-field pricing (overround spread) and not of the horse.",
      "rule": "last_pulled_up = 1 AND last_run_style IN ('held_up', 'midfield') AND field_size >= 10"
    },
    "pass_criteria": {
      "min_bets": 150,
      "min_ae": 1.1,
      "max_p": 0.05
    },
    "forward_bar": {
      "min_bets": 100,
      "min_ae": 1,
      "max_p": 0.2
    },
    "confounders": [
      "season: small fields cluster in the summer jumps period (after late April), the same window where the lone-PU lift sat, so field size and season are tangled and need a check by period",
      "going_group: summer ground is faster, which may suit horses that pulled up on soft winter ground",
      "sp_decimal: small fields shorten every price; the lift may sit in a band of prices, and 16/1+ has failed in every split",
      "race_class: summer cards are lower class, so a class drop may stand in for field size",
      "race-level pricing: overround per runner is different in small fields, which the same-race control picks up",
      "overlap with the long-gap split: horses coming back after a break in summer may be the same runners"
    ],
    "mechanism": "A held-up or midfield horse that pulled up last time may have been stopped before it was beaten into the ground. Today, in a small field (9 or fewer runners, common on summer jumps cards), it faces few rivals and a place or win is easier to get. The rival says the parent pool's lift (1.198) comes from this field-size effect and not from anything the horse held back. If so, the lift should sit in small fields and be missing in large ones.",
    "market_miss": "The P in the form makes the horse easy to lay. In small fields the overround is shared among few runners, and the market may not shorten a recent pull-up enough even though there are few genuine alternatives. The result is a price that is too long relative to its real chance in a thin field.",
    "prediction": {
      "direction": "positive",
      "size": "A/E 1.30+ on 350+ bets in fields of 9 or fewer, with the 10+ runner control near par",
      "falsified_if": "Small-field A/E at or below the parent's 1.198, or the 10+ runner control showing a lift as large as the small-field rule. Either would rule out field size as the carrier of the lift."
    },
    "notes": "The draft rule is kept: every column is pre-off, and last_run_style has been available since 2025-11-01 in GB and IRE. The key comparison is with the 10+ runner control. Even a positive result in small fields does not prove the field-size story unless it beats that control. Because season and field size are confounded, the falsification stage should split by period: before and after 27 April 2026."
  }

Timeline

1. Job 2336 finished

   17 Oct, 01:46 · engine:compute-1

2. H-d084cd23 Discovery → KILLED fails: A/E 1.09, needs 1.1; p 0.2823, needs 0.05 or lower; periods [0.95, 1.38], each needs 1.0 (not a niche pass: A/E 1.09 below 1.1)

   17 Oct, 01:46 · engine:evaluator

3. discovery result for H-d084cd23 (broad rule): fail

   17 Oct, 01:46 · engine:evaluator

4. H-d084cd23 Designed → RESEARCHdiscovery measuring on the discovery zone

   17 Oct, 01:46 · engine:orchestrator

5. compute-1 took job 2336 (evaluate, attempt 1)

   17 Oct, 01:46 · compute-1

6. Job 2334 finished

   17 Oct, 01:44 · engine:ai-1

7. Queued evaluate on the compute lane: discovery run of the broad rule

   17 Oct, 01:44 · engine:scheduler

8. H-d084cd23 Proposed → RESEARCHdesigned pre-spec written and hashed

   17 Oct, 01:44 · agent:designer@v1

9. ai-1 took job 2334 (design, attempt 1)

   17 Oct, 01:44 · ai-1

10. Queued design on the ai lane: write the pre-spec

    17 Oct, 01:43 · engine:scheduler

11. New idea H-d084cd23: Rival: held-up PU lift is a small-field summer artefact (rival-explanation)

    17 Oct, 01:43 · agent:investigator@v1

Every change is an event with an actor: an agent and its version, an engine, a human, or the importer. Current stage: Killed.